Andre Slavescu

Just to chime in, if this operator is implemented, when I eventually finish implementing conv3d, I can accelerate the operation for larger input kernel shapes. FFT conv tends to outperform...

Apple silicon is not supported. There exists a Metal plugin for Jax (which you would have to change out for the cuda jax build), but you will run into problems...

That just means you don't have the appropriate number of devices. The mesh_shape is the configuration for what is expected, in this case 8 devices to distribute the model over...

> @AndreSlavescu I have tried with 1. i did mesh_shape (1, 1) instead mesh_shape (1, 8) In theory that should work in terms of not breaking at that point, but...

> > > @AndreSlavescu I have tried with 1. i did mesh_shape (1, 1) instead mesh_shape (1, 8) > > > > > > In theory that should work in...

@WoosukKwon Hi sorry for the delayed reply, had a busy schedule this past week. I won't have much time to continue this coming week, so please continue on it if...

What driver version are you running?

It seems like you have not installed a GPU driver for your GPU. If you try running nvidia-smi, it will not work for this reason. Refer to this information page:...

> no CUDA-capable device is detected Okay in that case it seems you have installed it. There are quite a few problems if you install it a certain way, so...

I think it looks perfect. Maybe worth making a small PSA to move all general help questions to discussions as people are still posting in issues?